Self-Trained Target Detection of Radar and Sonar Images Using Automatic Deep Learning

Computational Intelligence and Neuroscience

Tang

Zhong

et al. 2022

Self Cite

The effectiveness of CycleGAN is demonstrated to outperform recent approaches for semisupervised semantic segmentation on public segmentation benchmarks. In contrast to analog images, however, the acoustic images are unbalanced and often exhibit speckle noise. As a consequence, CycleGAN is prone to mode-collapse and cannot retain target details when applied directly to the sonar image dataset. To address this problem, a spectral normalized CycleGAN network is presented, which applies spectral normalization to both generators and discriminators to stabilize the training of GANs. Without using a pretrained model, the experimental results demonstrate that our simple yet effective method helps to achieve reasonably accurate sonar targets segmentation results.

Section: Experiments and Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Spectral Normalized CycleGAN with Application in Semisupervised Semantic Segmentation of Sonar Images

Computational Intelligence and Neuroscience

Tang

Zhong

et al. 2022

Self Cite

“…Due to the specific properties, different machine learning models can have different results even facing the same input [ 4 ]. Compared with other neural networks [ 5 – 7 ], the convolutional neural network (CNN) has a better performance in the processing of image, including radar and sonar images, facial images, and hand gesture images [ 8 – 10 ]. Therefore, it also has been widely used in the recognition of radar waveforms [ 7 , 11 – 20 ].…”

Section: Introductionmentioning

confidence: 99%

LPI Radar Waveform Recognition Based on Neural Architecture Search

Computational Intelligence and Neuroscience

et al. 2022

Self Cite

In order to reach the intelligent recognition, the deep learning classifiers adopted by radar waveform are normally trained with transfer learning, where the pretrained convolutional neural network on an external large-scale classification dataset (e.g., ImageNet) is used as the backbone. Though transfer learning could effectively avoid overfitting, transferred models are usually redundant and might not generalize well. To eliminate the dependence on transfer learning and achieve high generalization ability, this paper introduced neural architecture search (NAS) to search the suitable classifier of radar waveforms for the first time. Firstly, one of the innovative technologies in NAS called differentiable architecture search (DARTS) was used to design the classifier for 15 kinds of low probability intercept radar waveforms automatically. Then, a method with an auxiliary classifier called flexible-DARTS was proposed. By adding an auxiliary classifier in the middle layer, the flexible-DARTS has a better performance in designing well-generalized classifiers than the standard DARTS. Finally, the performance of the classifier in practical application was compared with related work. Simulation proves that the model based on flexible-DARTS has a better performance, and the accuracy rate for 15 kinds of radar waveforms can reach 79.2% under the −9 dB SNR which proved the effectiveness of the method proposed in this paper for the recognition of radar waveforms.

“…In order to recognize randomly arbitrarily rotated patterns, CNNs have been expected to learn patterns in different orientations to obtain rotation invariance [5][6][7][8][9] , where the most simple but widely used method is rotation data augmentation 9,10 . Rotation data augmentation assigned the same category label to patterns and their rotated versions and let CNNs repeatedly learn them by rote at the training time.…”

Section: Introductionmentioning

confidence: 99%

Pre-rotation only at inference-time: a way to rotation invariance

Fourteenth International Conference on Digital Image Processing (ICDIP 2022)

Tang²,

Zhong³

et al. 2022

Weight sharing across different locations makes Convolutional Neural Networks (CNNs) space shift invariant, i.e., the weights learned in one location can be applied to recognize objects in other locations. However, weight sharing mechanism has been lacked in Rotated Pattern Recognition (RPR) tasks, and CNNs have to learn training samples in different orientations by rote. As such rote-learning strategy has greatly increased the difficulty of training, a new solution for RPR tasks, Pre-Rotation Only At Inference time (PROAI), is proposed to provide CNNs with rotation invariance. The core idea of PROAI is to share CNN weights across multiple rotated versions of the test sample. At the training time, a CNN was trained with samples only in one angle; at the inference-time, test samples were pre-rotated at different angles and then fed into the CNN to calculate classification confidences; at the end both the category and the orientation were predicted using the position of the max value of these confidences. By adopting PROAI, the recognition ability learned at one orientation can be generalized to patterns at any other orientation, and both the number of parameters and the training time of CNN in RPR tasks can be greatly reduced. Experiments show that PROAI enables CNNs with less parameters and training time to achieve state-of-the-art classification and orientation performance on both rotated MNIST and rotated Fashion MNIST datasets.